RE: [HACKERS] Cache query (PREPARE/EXECUTE)

> -----Original Message-----
> From: Karel Zak - Zakkr [mailto:zakkr(at)zf(dot)jcu(dot)cz]
>
> > Though current SPI stuff saves prepared plans to TopMemory
> > Context,we couldn't remove them forever. It seems that SPI
> > should also be changed in its implementation about saving
> > plans.
>
> Yes, I know about SPI plan saving... from here is my inspiration
> with TopMemoryContext. But we have in current PG code very often
> any cached queryPlan/Tree (PREPARE, SPI and Jan's RI saves plans
> to TopM. too), I agree with Tom that is not bad idea saving all
> plans to _one_ specific MemoryContext.
>
> My idea is make any basic routines for query cache (hash table,
> ExecuteCachedQuery() ...etc) and use these routines for more
> operation (SPI, FKeys, PREPARE..). Comments?
>
> > Note that freeObject() is unavailable at all.
> > We would be able to free PREPAREd resources by destroying
> > corrsponding memory context.
>
> If I good understand, we can't destroy any plan? We must

I think so. The problem is that Node struct couldn't be freed safely
due to the lack of reference count in its definition. As far as I see
plans could be destroyed only when the memory context in which
they are placed are destroyed.

> destroy _full_ memory context? If yes (please no), we can't
> make a DROP PLAN command or we must create for each plan specific
> memory context (and drop this specific Context (Jan's original idea)).
>

You can DROP a PLAN by removing its hash entry but of cource
there remains memory leak.

> If I call SPI_saveplan(), is the plan forever save in
> TopMemoryContext? (hmm, the SPI is memory feeder).
>

Probably.

Regards.

Hiroshi Inoue
Inoue(at)tpf(dot)co(dot)jp